It makes perfect sense. All it has are words. The thinking of an LLM is necessarily only as rich as the information content of the tokens preceding the next one.

All it takes is one buzzword, and then the likelihood of more appearing skyrockets (generally), because LLMs are (so far) most successful when their response is coherent with what came before.